This paper proposes an active contour algorithm for detecting objects in vector-valued images, such as RGB or multispectral images. The model is an extension of the scalar Chan–Vese algorithm to the vector-valued case. It minimizes a Mumford–Shah functional over the contour length and the sum of fitting errors across each component of the vector-valued image. The model can detect edges with or without gradient, and it robustly handles noise without requiring a priori denoising. The algorithm uses the level set method to determine the boundary of detected objects, allowing for automatic topology changes and handling of cusps and corners. Experimental results demonstrate the effectiveness of the model in detecting objects with missing parts in different channels, occlusions, and in noisy data, showing superior performance compared to classical methods.This paper proposes an active contour algorithm for detecting objects in vector-valued images, such as RGB or multispectral images. The model is an extension of the scalar Chan–Vese algorithm to the vector-valued case. It minimizes a Mumford–Shah functional over the contour length and the sum of fitting errors across each component of the vector-valued image. The model can detect edges with or without gradient, and it robustly handles noise without requiring a priori denoising. The algorithm uses the level set method to determine the boundary of detected objects, allowing for automatic topology changes and handling of cusps and corners. Experimental results demonstrate the effectiveness of the model in detecting objects with missing parts in different channels, occlusions, and in noisy data, showing superior performance compared to classical methods.